Journal Club
Currently our Journal Club takes place every Tuesday at 1 pm CEST. The meeting is in a hybrid format on Zoom and the conference room at the Maria-von-Linden Strasse 4th floor:
Topic: MLCS Journal Club
Time: Tuesdays, 1 pm CET
Join Zoom Meeting: Send email request
Past meetings
[31.05.22 - Canceled]
[24.05.22 - EGU General Assembly Meeting in Vienna]
[17.05.22 - EGU Test Talks]
10.05.22 - Thomas et al. (2018) The Role of Stochastic Forcing in Gernerating ENSO Diversity
03.05.22 - Di Capua et al. (2020) Dominant patterns of interaction between the tropics and mid-latitudes in boreal summer: causal relationships and the role of timescales
12.04.22 - Xception: Deep Learning with Depthwise Separable Convolutions
05.04.22 - Using network theory and machine learning to predict El Niño
29.03.22 - Universal gap scaling in percolation
22.03.22 - Sequential Neural Likelihood
15.03.22 - Towards Understanding Ensemble, Knowledge Distillation and Self-Distillation in Deep Learning
01.03.22 - Variational Inference with Normalizing Flows
22.02.22 - Variational Autoencoding of PDE Inverse Problems
15.02.22 - Auto-Encoding Variational Bayes
01.02.22 - Physics-informed semantic inpainting: Application to geostatistical modeling
25.01.22 - Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
18.01.22 - Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting
11.01.22 - Relational inductive biases, deep learning, and graph networks
14.12.21 - Zambrano et al., Prediction of drought-induced reduction of agricultural productivity … (2018)
07.12.21 - Kholodovsky et al., A generalized Spatio-Temporal Threshold Clustering method for identification of extreme event patterns (2021)
23.12.21 - Andersson et al., Seasonal Arctic sea ice forecasting (2021)
16.11.21 - Yuan et al., Typhoon Intensity Froecasting Based on LSTM (2021)
02.11.21 - Praveen et al., Analyzing trend and forecasting of rainfall changes in India (2020)
26.10.21 - Abatzoglou et al., Multivariate climate departures have outpaced univariate changes across global lands (2020)
19.10.21 - Ravuri et al., Skillful Precipitation Nowcasting using Deep Generative Models of Radar (2021)
12.10.21 - A quick guide through the IPCC 2021
05.10.21 - Cleary et al., Calibrate, emulate, sample, Jour. Comp. Phys. (2021)
28.09.21 - Capotondi et al., The Nature of the Stochastic Wind Forcing of ENSO (2018)
21.09.21 - Deo et al., Drought forecasting in eastern Australia … (2017)
14.09.21 - Chattopadhyay et al., Analog Forecasting of Extreme-Causing Weather Patterns Using Deep Learning (2020)
07.09.21 - Kretschmer et al., Using Causal Effect Networks to Analyze Different Arctic Drivers of Midlatitude Winter Circulation (2016)
31.10.21 - Giffard-Roisin et al., Tropical Cyclone Track Forecasting using Fused Deep Learning from Aligned Reanalysis Data (2020)
24.10.21 - Vicedo-Cabrera et al., The burden of heat-related mortality attributable to recent human-induced climate change (2021)
17.10.21 - Weyn et al., Sub-Seasonal Forecasting With a Large Ensemble of Deep-Learning Weather Prediction Models (2021)
10.08.21 - Liu et al., Stein Variational Gradient Descent (2016)
03.08.21 - Walker et al., Interannual Variability in the Large-Scale Dynamics of the South Asian Summer Monsoon (2015)
27.07.21 - Gerhardus, Runge (2020) High-recall causal discovery for autocorrelated time series with latent confounders
20.07.21 - Sinha et al. (2020) Variational Autoencoder Anomaly-Detection of Avalanche
06.07.21 - Runge et al. (2019) Detecting and quantifying causal associations in large nonlinear time series datasets
22.06.21 - Cachay et al. (2021) The World as a Graph: Improving El Nino Forecasts with Graph Neural Networks
15.06.21 - Mokhov et al. (2011) Alternating mutual influence of ENSO and Indian Monsoon
08.06.21 - Traxl et al. (2016) Deep Graphs
01.06.21 - Sun et al. (2014) Monthly streamflow forecasting using Gaussian Process Regression
25.05.21 - Liang et al. (2020) Gated Recurrent UnitNetwork for Wind Speed Forecasting
18.05.21 - Pathak et al. (2018) Model-Free Prediction of Large Spatiotemporally Chaotic Systems from Data: A Reservoir Computing Approach
04.05.21 - Bécenac et al. (2019) Deep learning for physical processes: incorporating prior scientific knowledge
13.04.21 - Domeisen et al. (2020) The surface impact of sudden stratospheric warming events
06.04.21 - Li et al. (2020) Impact of NAO and atmospheric blockin on European heatwaves
30.03.21 - Yang et al. (2020) Poleward Shift of the Major Ocean Gyres Detected in a Warming Climate
23.03.21 - England et al. (2020) Tropical climate responses to projected Arctic and Antarctic sea-ice loss
16.03.21 - Acosta et al. (2019) Competing Topographic Mechanisms for the Summer Indo‐Asian Monsoon
12.03.21 - Oliver et. al (2020) Marine Heatwaves
23.02.21 - Literature review on reduced representations
16.02.21 - Literature review on river discharge
09.02.21 - Literature review on ITCZ prediction
02.02.21 - Literature review on Global monsoon
26.01.21 - Literature review on ENSO Diversity
19.01.21 - Johnson et al. (2019) SEAS5: the new ECMWF seasonal forecast system
12.01.21 - Yan et al. (2020) Exploring the ENSO Impact on Basin‐Scale Floods
05.01.21 - Literature review of Normalizing Flows
15.12.20 - Guo et al. (2017) Identify distinct Patterns of Tropical Pacific SST Anomalies using SOMs
22.12.20 - Peixoto (2019) Bayesian Stochastic Blockmodeling
08.12.20 - Hegerl et al. (2019) Causes of climate change over the historical record
01.12.20 - Yang et al. (2019) Compensatory climate effects link trends in global runoff to rising atmospheric CO concentration
24.11.20 - Pante et al. (2020) Resolving Sahelian thunderstorms improves mid-latitude weather forecasts
17.11.20 - Wang et al. (2019) Diversity of the Madden-Julian Oscillation
10.11.20 - DiNezio (2020) Emergence of an equatorial mode in the Indian Ocean
03.11.20 - Hoskins (2020) The detailed dynamics of the June–August Hadley Cell
28.10.20 - Ding (2005) Circumglobal Teleconnection in the NH Summer
27.10.20 - Papagiannopoulou (2018) Hydro-climatic biomes via multitask learning
13.10.20 - Schneider et al. (2014) Migrations and dynamics of ITCZ
05.10.20 - Deser et al. (2017) ENSO variability
28.09.20 - Talk: Climate Change are we up for the Challenge
22.09.20 - Presentation of research ideas
15.09.20 - Houze et al. (2015) TRMM dataset summary
09.09.20 - Datasets, benchmarks and scores
04.09.20 - Rasp et al. (2020) Data-driven medium range weather forecasting
03.09.20 - Rasp et al. (2018) NN for learning subgrid proecesses
02.09.20 - Renard et al. (2019) Hidden Climate Indices for Floodings
01.09.20 - Petersik et al. (2020) Probabilistic ENSO Forecasting
14.12.21
Zambrano et al., Prediction of drought-induced reduction of agricultural productivity in Chile from MODIS, rainfall estimates, and climate oscillation indices (2018)
The paper was presented by Davide L.
07.12.2021
Kholodovsky et al. (2021) A generalized Spatio-Temporal Threshold Clustering method for identification of extreme event patterns
The paper was presented by Julia H.
30.11.21
Better Better — machine learning for improved climate models and projections - AI for Good
Watch AI for Good video.
23.12.2021
Andersson et al., Seasonal Arctic sea ice forecasting with probabilistic deep learning, Nature Communications, 2021
The paper was presented by Alexej O.
16.11.2021
Yuan et al., Typhoon Intensity Froecasting Based on LSTM (2021)
The paper was presented by Ranganatha B.
02.11.2021
Praveen et al., Analyzing trend and forecasting of rainfall changes in India using non-parametrical and machine learning approaches (2020)
The paper Praveen et al., Analyzing trend and forecasting of rainfall changes in India using non-parametrical and machine learning approaches (2020) was presented by Felix O.
Notes:
- non-parametric trend analysis
- mann-kendall-test with pre-whitening
- pettitt change point test
26.10.2021
Abatzoglou et al., Multivariate climate departures have outpaced univariate changes across global lands (2020)
The paper Abatzoglou et al., Multivariate climate departures have outpaced univariate changes across global lands (2020) was presented by Moritz H.
19.10.2021
Ravuri et al., Skillful Precipitation Nowcasting using Deep Generative Models of Radar (2021)
The paper Ravuri et al., Skillful Precipitation Nowcasting using Deep Generative Models of Radar (2021) was presented by Jannik K.
12.10.2021
A quick guide through the IPCC 2021
The quick guide through the IPCC 2021 was presented by Felix S.
The presentation can be found here.
05.10.2021
Cleary et al., Calibrate, emulate, sample, Jour. Comp. Phys. (2021)
The paper Cleary et al., Calibrate, emulate, sample (2021) was presented by Jakob S.
The presentation can be found here.
Discussion:
- How does ensemble Kalman inversion scale with number of parameters?
- What constraints do time-average have on the data and model?
- Can the method be extended to capture structural model biases?
28.09.2021
Capotondi et al., The Nature of the Stochastic Wind Forcing of ENSO (2018)
The paper Capotondi et al., The Nature of the Stochastic Wind Forcing of ENSO (2018) was presented by Bedartha G.
21.09.2021
Deo et al., Drought forecasting in eastern Australia … (2017)
The paper Deo et al., Drought forecasting in eastern Australia using multivariate adaptive regression spline, least square support vector machine and M5Tree model (2017) was presented by Davide L.
14.09.2021
Chattopadhyay et al., Analog Forecasting of Extreme-Causing Weather Patterns Using Deep Learning (2020)
The paper Chattopadhyay et al., Analog Forecasting of Extreme-Causing Weather Patterns Using Deep Learning (2020) was presented by Altaf A.
07.09.2021
Kretschmer et al., Using Causal Effect Networks to Analyze Different Arctic Drivers of Midlatitude Winter Circulation (2016)
The paper Kretschmer et al., Using Causal Effect Networks to Analyze Different Arctic Drivers of Midlatitude Winter Circulation (2016) was presented by Alexej O.
31.10.2021
Giffard-Roisin et al., Tropical Cyclone Track Forecasting using Fused Deep Learning from Aligned Reanalysis Data (2020)
The paper Giffard-Roisin et al., Tropical Cyclone Track Forecasting using Fused Deep Learning from Aligned Reanalysis Data (2020) was presented by Ranganatha B.
24.10.2021
Vicedo-Cabrera et al., The burden of heat-related mortality attributable to recent human-induced climate change (2021)
The paper Vicedo-Cabrera et al., The burden of heat-related mortality attributable to recent human-induced climate change (2021) was presented by Julia H.
17.10.2021
Weyn et al., Sub-Seasonal Forecasting With a Large Ensemble of Deep-Learning Weather Prediction Models (2021)
The paper Weyn et al., Sub-Seasonal Forecasting With a Large Ensemble of Deep-Learning Weather Prediction Models (2021) was presented by Felix O.
10.08.2021
Liu et al., Stein Variational Gradient Descent (2016)
The paper Liu et al., Stein Variational Gradient Descent (2016) was presented by Jannik T.
03.08.2021
Walker et al., Interannual Variability in the Large-Scale Dynamics of the South Asian Summer Monsoon (2015)
The paper by Walker et al., Interannual Variability in the Large-Scale Dynamics of the South Asian Summer Monsoon (2015) was presented by Felix S.
27.07.2021
Gerhardus, Runge (2020) High-recall causal discovery for autocorrelated time series with latent confounders
The paper Gerhardus, Runge (2020) High-recall causal discovery for autocorrelated time series with latent confounders was presented by Moritz H.
The presentation can be found here.
20.07.2021
Sinha et al. (2020) Variational Autoencoder Anomaly-Detection of Avalanche
The paper Sinha et al. (2020) Variational Autoencoder Anomaly-Detection of Avalanche was presented by Jakob S.
The presentation can be found here
06.07.2021
Runge et al. (2019) Detecting and quantifying causal associations in large nonlinear time series datasets
The paper by
Runge et al. (2019) Detecting and quantifying causal associations in large nonlinear time series datasets
was presented by Bedartha G.
The presentation can be found here
22.06.2021
Cachay et al., The World as a Graph: Improving El Nino Forecasts with Graph Neural Networks (2021)
The paper Cachay et al., The World as a Graph: Improving El Nino Forecasts with Graph Neural Networks (2021) was presented by Christian F.
Discussion:
- Overall an very interesting paper connecting climate networks with forecasting
- The learned graphs do not have a lot in common with climate networks and thus their interpretability is hard
- How would a saliency map of the GNN look like?
- As most statistical ENSO models it fails to predict the extreme events
The presentation can be found here
15.06.2021
Mokhov et al., Alternating mutual influence of ENSO and Indian Monsoon (2011)
The paper Mokhov et al., Alternating mutual influence of ENSO and Indian Monsoon (2011) was presented by Alexej O.
The presentation can be found here
08.06.2021
Traxl et al., Deep Graphs (2016)
The paper Traxl et al., Deep Graphs (2016) was presented by Julia H.
The presentation can be found here
01.06.2021
Sun et al., Monthly streamflow forecasting using Gaussian Process Regression (2014)
The paper
Sun et al., Monthly streamflow forecasting using Gaussian Process Regression (2014)
was presentd by Markus D.
The presentation can be found here
25.05.2021
Liang et al., Gated Recurrent UnitNetwork for Wind Speed Forecasting (2020)
The paper Liang et al., Gated Recurrent UnitNetwork for Wind Speed Forecasting (2020) was precented by Lea E.
The presentation can be found here
18.05.2021
Pathak et al. (2018) Model-Free Prediction of Large Spatiotemporally Chaotic Systems from Data: A Reservoir Computing Approach
The paper Pathak et al., Model-Free Prediction of Large Spatiotemporally Chaotic Systems from Data: A Reservoir Computing Approach (2018) was presentd by Felix S.
The presentation can be found here
04.05.2021
Bécenac et al. (2019) Deep learning for physical processes: incorporating prior scientific knowledge
The paper
Bécenac et al. (2019) Deep learning for physical processes: incorporating prior scientific knowledge
was presented by Jakob S.
The presentation can be found here
13.04.2021
Domeisen, Grams, & Papritz (2020) The role of North Atlantic–European weather regimes in the surface impact of sudden stratospheric warming events, Weather Clim. Dyn.
The paper Domeisen, Grams, & Papritz, Weather Clim. Dyn., (2020) was presentd by Bedartha G.
The presentation can be found here
06.04.2021
Li et al. (2020) Impact of NAO and atmospheric blockin on European heatwaves
The paper Li et al. (2020) was presented by Julia H.
The presentation can be found here
Discussion:
- probabilistic analysis would be more expressive P(EB|NAO) and P(EB|~NAO)
30.03.2021
Yang et al. (2020) Poleward Shift of the Major Ocean Gyres Detected in a Warming Climate
The paper by Yang et al. (2020) was presentd by Markus D.
The presentation can be found here
23.03.2021
England et al. (2020) Tropical climate responses to projected Arctic and Antarctic sea-ice loss
The paper
England et al., Tropical climate responses to projected Arctic and Antarctic sea-ice loss (2020)
was presented by Lea E.
16.03.2021
Acosta et al. (2019) Competing Topographic Mechanisms for the Summer Indo‐Asian Monsoon
The paper Acosta et al. 2019: Competing Topographic Mechanisms for the Summer Indo‐Asian Monsoon was presented by Felix S.
Acosta et al. studies the effect of topography on the Indo-Asian Monsoon. The presentation can be found here
Discussion:
- Typical monsoon behavior (wind and precipitation) is present even when the topography are removed. This emphasizing the eddy-driven mechanism theory.
- Iranian platau seems to have the greatest influence on the Equivalend potetial temperature pattern
- Westerly winds in the winter might be an important driver
12.03.2021
Oliver et. al - Marine Heatwaves (2020)
The paper Oliver et. al, Marine Heatwaves, Annual Review of Marine Science (2020) was presented by Jakob S. The presentation can be found here
23.02.2021
Literature review on reduced representations
The review was presented by Bedartha G. and the presentation can be found here.
Notes:
- dimensionality reduction techniques represent high dimensional data by a smaller set of essential features
- Why to use dimensionality reduction:
- reduce computational complexity
- reduce informational complexity, i.e. increase interpretability
- remove noise
- Dimensionality reduction methods:
- PCA, LLE, LEM, MDS, ISOMAPS can all be described as special forms of kPCA
- Other non-convex DR methods: Graph Clustering, VAE, SOM, NNMF
16.02.2021
Literature review on river discharge
Review was presented by Markus D.
09.02.2021
Literature review on ITCZ prediction
Review was presented by Lea E.
02.02.2021
Literature review on the Global monsoon
Review was presented by Felix S.
26.01.2021
Literature review to ENSO Diversity
Review was presented by Jakob S.
19.01.2021
** Johnson et al. (2019) SEAS5: the new ECMWF seasonal forecast system **
The paper by Johnson et al., Geosc. Model Dev., 2019 was presented by Bedartha G.
The presentation can be found here.
Discussion:
- SEAS5 seem to focus on the ENSO improvement but leads to a poorer description of the Indian Ocean in comparison to SEAS4
- SEAS5 ensemble prediction of the NAO does not capture the varibility of NAO. This suggests that SEAS5 poorly forecasts North America and Europe
- SEAS5 does not capture extratropical variability well. Here, a comparison to the Souther Hemisphere variability indices would have been nice.
- In summary, SEAS5 does forecast the Tropics, specifically the ENSO fairly well but lacks forecasting Europe.
12.01.2021
** Yan et al. (2020) Exploring the ENSO Impact on Basin‐Scale Floods**
The paper by Yan et al., Geo. Res. Lett., 2020 was presented by Markus D.
The presentation can be found here.
Discussion:
- water runoff and precipitation are not directly correlated. Soil properties, temperature and other factors play an important role
- Can we think of the monsoon as a symmetric phenomena? The south American monsoon region show maximum flood intensity in Nov (at the beginning of the monsoon season) and maximum flood frequency in March (at the end of the monsoon season)
05.01.2021
** Overview of Normalizing Flows **
An overview of Normalizing flows were presented by Lea E.
22.12.2020
** Peixoto (2019) Bayesian Stochastic Blockmodeling **
The book chapter by Peixoto, Adv. Net. Clust. and Blockmod., 2019 was presented by Felix S.
The presentation can be found here.
Summary:
- stochastic blockmodeling (SBM) for finding groups within networks
- number of clusters is obtained by minimizing the discription length
- graph-tool library indcludes all described methods and tools in the paper
- SBM can be generalized to weighted networks
15.12.2020
Guo et al. (2017) Distinct Patterns of Tropical Pacific SST Anomaly and their Impacts on the North American Climate
The paper by Guo et al., Journal of Climate, 2017 was presented by Jakob S.
The presentation can be found here.
Discussion:
- Self-organizing maps are another tool for dimensionality reduction
- this was mainly a methods paper
- biases of model ensembles are not discussed
- missing comparison to normal conditions
- datacube classification might be interesting
08.12.2020
Hegerl et al. (2019) Causes of climate change over the historical record
The paper by Hegerl et al., Environ Res Lett, 2019 was presented by Bedartha G.
The presentation can be found here.
Key points:
- Review paper on Climate change drivers
- Main Question: What factors caused decadel and multidecadel deviations from the greenhouse warming trend?
- Fingerprints: terminology of factors impacting global temperature e.g. solar fingerprint
- Global temperature can grouped into different periods of global trends
- Warming are not uniform over the globe
- Anthropogenic aerosol hamperes the GHG warming, especially in between 1950 and 1980
- Diurnal: temperature difference between max and min within a day
- Heat flux in the Atlantic is correlated to the AMOC
Discussion:
- Aerosols seem to play an important role for global warming which is not yet fully understood
- The paper focusses mainly on the Northern Hemisphere and Atlantic Ocean although discussing global climate change
- Effect of ENSO, PDO and Southern Ocean are not considered
01.12.2020
Yang et al. (2019) Compensatory climate effects link trends in global runoff to rising atmospheric CO concentration
The paper by Yang et al. (2019) was be presented by Markus D.
The presentation can be found here.
Summary:
Yang et al. analyzes the role of carbon-nitrogen (CN) cycle to the water runoff model from Joint UK Land Environment Simulator (JULES). They show that including the CN cycle improves the prediction of runoff trends on a local scale but have only a minor effect globally .
Discussion:
- actual runoff values are not presented
- actual runoff value comparison is not possible since it is not measurable
- runoff models do likely not capture actual river discharge
24.11.2020
Pante et al. (2020) Resolving Sahelian thunderstorms improves mid-latitude weather forecasts
The paper Pante, Knippertz, Nature Comms, 2020 was presented by Lea E.
The presentation can be found here.
Summary:
Pante et al. show that explicitly resolving summer time Sahelian mesoscale convective systems in a novel two-way nesting apporach in the ICON numerical weather prediction model impacts on forcast biases over West Africa. This also improves forecasts in the extratropics and Europe since the Saharan heat low (SHL), tropical easterly jet (TEJ) and african eastery jet (AEJ) quickly carry signals out of Africa.
Discussion:
- interesting perspective on teleconnections: improve the model somewhere in the world leads to an improvement of the forecast in another part of the world
- understanding teleconnections from first principle is difficult
- cloud parametrization on a global scale might not be necessary to improve weather forecasts since only some regions play an important role
- model nesting: placing higher resolution domain within a coarser domain
- two-way nesting: compute both domains at the same time
17.11.2020
Wang et al. (2019) Diversity of the Madden-Julian Oscillation
Wang et al., Sci Adv, 2019 presented by Felix S.
The presentation can be found here.
Discussion:
- MJO is the strongest intraseasonal variability
- How can we conclude the effect of Kelvin waves from wind vector fields?
10.11.2020
DiNezio (2020) Emergence of an equatorial mode in the Indian Ocean
DiNezio et al., Sci Adv, 2020 to be presented by Jakob S.
The presentation can be found here.
Summary:
Methods:
- Numerical simulations of past and future climate changes
- Coupled Model Intercomparison Project 5 (CMIP5) run under “buisness as usual” high emission scenario
Only some models get the SST gradient changes in the IO correct
- Analyze simulations of the Last Glacial Maximum (LGM) around 21000 years ago (CESM1)
Results:
- IO normal conditions become more similar to the Pacific and Atlantic with westerly winds and shallower thermocline in the eastern IO
- CMIP5 models predict increasing westerly winds with GHG
- Under greenhouse warming and LGM the SST variability increases in the eastern equatorial IO (EEIO)
- This could lead to EEIO mode like in Atlantic and Pacific
- Causation of the EEIO mode:
- Equatorial winds become more easterly
- Increased upwelling and eastward shoaling thermocline in JAS
- Self enhancing effect known as Bjerknes atmospheric-oceanic feedback -> Development of SSTAs in EEIO in Aug-Sep-Oct months
- EEIO SST variablity is destingued from IOD, which is shown by disabled ENSO and IOD mode in LGM sim.
(disable ENSO and IOD means setting Nino3.4 and IOD index to climatology)
- EEIO mode also shows the negative/cold phase
Impacts for the second half of the century:
- equatorial mode would drive rainfall variability with stronger amplitude
- rainfall deficits (droughts) over the Horn of Africa as well as Southern India
- increased rainfall over Indonesia and Northern Australia
- so far IOD had not as strong impacts because of its strength
- CMIP5 predict 2-4 equatorial IO modes per decade
Discussion:
- Why do only some CMIP5 models show this equatorial mode emergence?
- Is there some reason for this deviation?
- Given the CMIP5 models how likely would such an equatorial mode emerge? What are the uncertainties?
03.11.2020
Hoskins (2020) The detailed dynamics of the June–August Hadley Cell
The paper Hoskins 2020 was presented by Bedartha G.
Summary:
Hoskins et al. examine the Hadley cell in JJA by observing the angular
momentum and vorticity. They show,that that tropical convections and
meridional circulations drive the Hadley Cell movement. Thus, the main
contribution to the summer (JJA) Hadley cell is from the Indian Ocean
and the west Pacific. Equatorial air movement from southern to northern
Hemisphere occurs in filaments in the upper troposphere.
Observation variables:
- angular momentum
- absolute vorticity: vorticity with added earths rotation
- potential vorticity: vorticity on an isotropic surface
Take home message:
- Hadley cell in summer is really only one
- Hadley cell is not at all even around the globe (longitudes)
- vorticity is strongly related to convection
28.10.2020
Ding (2005) Circumglobal Teleconnection in the Northern Hemisphere Summer
Felix S. presented the paper presentation:
Ding and Wang show evidence that teleconnections in the Northern Hemisphere are linked with each other during boreal summer and on interannual variability. They term this connection as circumglobal teleconnection (CGT) pattern.
- geopotential height: the height of one particular pressure, this is an intuitive variable for atmospheric flow
- teleconnections their definition: relationship in the low-frequency variability by studying EOFs and look at their spatial regions
- circumglobal teleconnection is defined by positive correlation of maximum standard deviation of geopotential height, the maximum over India is used as an index CGTI
- the second EOF of geopotential height shows hight correlation to the CGT
- CGTI is correlaed to Indian Summer Monsoon but not to ENSO, the correlation between CGTI and ENSO is via ISM
- two scenarios qualitatively explain the CGT:
- anomaly height due to monsoon is transported by rossby waves
- anomalous height above europe which is transported by rossby wave which gives a height above india
Discussion:
- connection between teleconnections defined as low-frequency
variability of EOFs and climate network teleconnections is not fully
clear [paper by Donges]
- fuzzy network teleconnections may be similar to EOF teleconnections
- both definitions are based on the correlation matrix
- CGT definition has got rising attention in 2020
27.10.2020
Papagiannopoulou (2018) Global hydro-climatic biomes identified via multitask learning
Markus D. presented the paper Presentation:
Aim: develope data driven approach to quantify the response of vegetation to local climate variables
Discussion:
- comparing Granger causalities (like in Fig. 3) should be done using some statistical test
- including climate features in the multi task learning model increases prediction stronger than in the single task learning model
- log scale for precipitation in the scatter plot would be better
- this method is well suited for high dimensional input clustering where standard methods, e.g. k-means, etc. are weak
13.10.2020
Schneider (2014) Migrations and dynamics of the intertropical convergence zone
Presentation by Lea:
- Give a definition of ITCZ
- analysis of ITCZ migrates to warmer hemisphere
- energy flux are weak near the ITCZ
- definition of ITCZ by divergence of moist energy flux
- annual changes of ITCZ in the Indian Ocean is much stronger than in the Pacific and Antlantic which can’t be explained by the energy flux
- ITCZ is found where the energy flux changes sign
- moist static energy: the heat released by a air parcel when moved from some height to the upper atmosphere, i.e. the heat released when water vapour condensates (indicator of water vapour in an air parcel)
- model of div F = S - F - O
- O, F, S are measured
- this allows to obtain the energy flux equator = ITCZ
- northern shifted ITCZ over Pacific and Atlantic is due to the AMOC
- in the Indian Ocean the monsoon is likely to have an impact on ITCZ position
- From La Nina to El Nino the ITCZ moves southwards which is counter
intuitive
- explanation from ocean uptake which significantly decreases
- greenhouse warming in the Northern Hemisphere is stronger than in the Southern Hemisphere which leads to a movement of ITCZ to the NH
Discussion:
- moist static energy heat flux is obtained by an weighted integral of the wind fields over the pressure (weighted by the value of moist energy values)
- the ITCZ influences the monsoon, its likely not the other way around
- the second maxima of precipitation south of the Equator in the Indian Ocean in JJA might be interesting to observe (Fig. 2 b)
05.10.2020
Deser et al. 2017
https://doi.org/10.1175/JCLI-D-16-0844.1
Jakob S. presented the paper Presentation Deser (2017)
Discussion:
- averaging already the DJF months loses a lot of information about the variability
- the assumption of interchangable ENSOs is not necessarily true and should be tested
- constructing all possible pairs of El Ninos and La Ninas and sample from them would possibly give a larger uncertainty (this would go beyond the interchangable ENSO)
28.09.2020
Talk: Climate Change are we up for the Challenge
Youtube lecture Brian Hoskins - Climate Change are we up for the Challenge
Notes:
- the earth without an atmosphere has temperature of -18
- with greenhouse gases its 15 degree. The greenhouse gases are water vapour, carbon dioxide, methane
- Carbon dioxide measurements:
- 100t of CO_2 emitted now (2016)
- in 1000 years without any more emission 15 - 40 t will remain
- Global warming is not equally distributed, Northern hemishpere is stronger affected
- sea level change due to ice melting and water expansion
- no sea ice in summer in the mid century
- warmer atmosphere can hold more water thus, the number of extreme rain events increase (Clausius–Clapeyron relation)
- cyclones become stronger because warmer oceans give more energy to the winds
- Projections to the future 2081 - 2100 by 4 degree rise of GST (IPCC 2013)
- 11 degrees warmer over the landmass in the Northern Hemisphere
- land ocean contrast would yield to new weather
- 1 m rise of water level
- Oceans change pH value
- Compensating Methods:
- Getting CO2 out of atmosphere
- Geoengineering
- Mitigation
- area under the curve as budget\
- Challenge to society and politics
- Can we look beyond our own group
- CAn we look beyond the short term
- Number/Measure for communication, number of aircondition, max temperatures
22.09.2020
Presenting Research Ideas
Felix and Jakob where presenting a set of possible research ideas.
- Felix’s idea was related to the global monsoon
- Jakob’s idea was related to ENSO complexity prediction.
15.09.2020
Houze 2015: The variable nature of convection in the tropics and subtropics A legacy of 16 years of the Tropical Rainfall Measuring Mission satellite
https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2015RG000488
Summary of precipitation and convection patterns over the globe which are obtained by precipitation radar of the Tropical Rainfall Measuring Mission (TRMM). Convection systems have different types which cannot be viewed as a single kind of entity.
Discussion:
- GPM dataset is an extension of the TRMM dataset which reaches further to the poles
- TRMM dataset on precipitation is thus preprocessed
- Stratiform and convective clouds are different in their vertical latent heat profile, shallow and tall, respectively
- Convection systems and thus extreme rain might have different physical origin (DCC or WCC)
- Traxl2016 network in space time grouping extreme events
- Are extreme rainfall teleconnections different for these convection classes?
- Idea: Which features in terms of precipitation (e.g. extreme rainfall) corresponds to DCC and WCCs
- Idea: Scatter plot of p(DCC) and p(WCC) to see cluster and project them to real space instead of comparing maps
09.09.2020
Datasets, Benchmarks, Scores and Metrics
Benchmarks: https://www.overleaf.com/8663366341vrwmtgrqrxks
- benchmark for algorithms
- algorithm should not over or under fit
- key question (of a scientific domain) with respect to a particular algorithm
- benchmark for datasets/problem
- Properties: domain specific problem, not easily solvable, naive solution is not satisfying, baseline or ground truth
- key question of a scientific domain which is closely related to a specific dataset
- [Uphoff2020] website to store benchmark climate datasets
- Problems/Tasks are benchmarks which are either related to datasets or algorithms
- Api/features of the benchmark task should be formulated from the domain scientists (quantitative quantifiers should be left open)
- In climate science: tasks are usually with respect to dataset where the new approach/model should be better than existing methods
- Dimensions of quantifiers: computational speed and memory, accuracy, scale ability
Scores: Prediction scores and metrics
- idea to sort metrics or errors by their property, i.e. Markov property for time-series
- significance tests are only used by statisticians (literature from Alvarez)
- ACC:
could be inferred from the data by using a bayesian structural model
- idea to use relational-databases to sort tasks, datasets, papers and scores (notion.so, airtable)
Research Ideas:
- Formulate: research question, dataset and method
- think about possible results (hypothesis)
- make a timeline/plan including bottlenecks
04.09.2020
Rasp, Thuerey 2020: Purely data-driven medium-range weather forecasting achieves comparable skill to physical models at similar resolution
https://arxiv.org/pdf/2008.08626.pdf
Discussion:
- input data 240 000 values at each time point (maybe three time steps)
- course grid of appox. 360km
- ERA5 has a resolution of 0.25 degree (you can download even finer grid from ERA, probably interpolated)
- running average over the year and plot variance
- seasonality may also play a role (same apply for long range phenomena, e.g. ENSO, where other periodic phenomena play a role)
- gradients are in respect to time (i.e. sample direction)
- median would be more expressive to capture extreme events
Idea which came up:
Course grain the input at the lat/lon nodes by a correlation matrix to create an adjacency matrix. This would reduce dimension to relevant correlations to before feeding into an ML algorithm for e.g. ENSO prediction. This would reduce the input to relevant patterns that any ML approach must anyway recognize.
Ideas how to proceed are:
- set a value threshold in adjacency matrix to reduce input data for each lead-time
- do this for a set of input variables (SST, precipitation, windspeed at different pressure levels…)
- combines climate networks with learning algorithms
- Compare to NN which gets not processed input data. NN might learn the same thing. Either way would be interesting.
- Rheinwalt2015
03.09.2020
Rasp 2018: Deep learning to represent subgrid processes in climate models
https://www.pnas.org/content/115/39/9684
The computational expensive cloud resolving model, SPCAM, is surrogated by training a deep neural network (NNCAM).
- Once trained, the NNCAM runs 20 times faster
- The NNCAM captures the mean climate well
- NNCAM learned to conserve energy
- The network is able to interpolate between different temperature scenarios but fails on extrapolation
- NNCAM outperforms a parametrized surrogate model, CTRLCAM
Method:
- NN with 9 layers and 256 nodes per layer
- dimension of data point: x=94, y=65 with 140 mio training points
Question and Discussion:
- CTR is the control/base line model
- pressure is exponential height
- related field: using NN for downscaling, i.e. form course to fine scale
- related work: intra day extreme precipitation Mishra 2018
02.09.2020
Renard 2019: Hidden Climate Indices from Occurances of Hydrologic Extremes
Develop a probabilistic model with a set of latent variables which is used to describe temporal and spatial extreme precipitation. The learned latent variables are called hidden climate indices (HCIs) which show correlation to climate variables in some cases.
- avoid relying on standard climate indices, which might be poor predictors
- showing method on synthetic case, Australian coast line (ENSO dependent) and floods in France
Method:
- Occurrences of event are described by Bernoulli distribution
- spatial parameters are multivariate Gaussian distributions
- conditional independence between parameters in space and time
- stepwise Bayesian inference using adaptive MCMC
- similar method that probabilistic PCA
Case studies and results:
- Synthetic data are well described by model even with missing data, standard effect of hidden variables show significance
- Eastern Australian spring floods are described by K=3 HCIs, where the first HCI correlates with the NINO4 index
- Autumn floods in France are also well described by the model with K=6 HCIs. The HCIs do not correlated to standard climate indices. The first index slightly correlates with large convection in the north of France and strong winds in the south of France.
Improvements:
- include interactions between parameters of HCIs
- go beyond linear time or space behaviour
Discussion:
- comparison to correlation network for teleconnections
- how many parameters to choose
- PDO and NAO are indices which result of PCA
- ONI is a three month average of Nino3.4, adapted ONI with thirty year average of temperature
- main message: creating time series from discrete occurance extreme events
- extreme value distributions: maximum of each block in a time series (used in hydrology for the worst floods)
01.09.2020
Petersik 2020: Probabilistic forecasting of El Nino using NN Models
ENSO forecasting using Gaussian density neural networks (GDNN) and Quantile Regression Neural Networks (QRNN).
- GDNNs outperform QRNN because prior assumptions on distribution compensate for small dataset
- high prediction skill by using few variables (ONI, WWV, DMI, zonal wind stress anomaly, metric of ECNS)
Model:
- GDNN: ensemble of multilayer perceptrons with mean and variance output nodes
- QRNN: predicts the location of a quantile
- Regularization by dropout and early stopping
- output of the models are the ONI for lead time
Predictor Variables:
- Oceanic Nino Index (ONI): 3 month running average of SST anomaly in NINO 3.4 region
- WWV: volume of water at 20(^\circ)C isotherm
- Dipole Mode Indes (DMI): difference between the monthly area-averaged SSTA of the western and southeastern equatorial Indian Ocean
- zonal wind stress anomaly in the western Pacific
- ECN
Results:
- Anomaly correlation coefficient (ACC): Pearson correlation between predicted mean and observed ONI
- Quantile skill score (QSS): positive QSS indicates that the model is better than the reference
- For El Nino time periods ACC have weak seasonal variation in contrast to La Nina
- GDNN outperforms QRNN in ACC and QSS
Questions and Critics:
- importance of parameters for the prediction (heat map)
- evaluation of input variables, e.g. start with 20 adhoc predictors and reduce them by tracking prediction quality
- weak discussion of uncertainties
- correlation ACC do not predict absolute values of ONI
- humming distance: normalized missing links in network
- La nina is usually better predicted than El nino, which is in contrast to Fig. 2